1. Antecedentes generales
3.1 Diversas teorías sobre las “Dificultades de Aprendizaje”
Structural equation models continually make causal assumptions. Before going into a detailed discussion of a series of subsets of that will culminate in the final model of the marketing research orientation of tourism organisations, it is necessary to discuss briefly the nature of causality, the condition of causation, and the limits of causal modelling.
The general definition of causality as used in structural equation modelling states that if a change in one variable (γ1) accompanies a change in another variable (χ1) then χ1 is a cause of γ1, provided the latter is isolated from all other influences. The definition of cause has three components: isolation, association, and the direction of influence. As correlation does not imply causation, association between two variables is not enough by itself. Isolation must come before association. Then there is the problem of establishing direction – that the association is due to χ1 affecting γ1, and not the other way around.
Holland (1986) argued that a variable could be a cause only if it can be subject to human manipulation. There are problems with this view as it is goes against common
sense and intuition: Earthquakes do cause destruction of buildings and other property, and the moon does cause the tides. This thesis accepts Bollen’s (1989, p. 41) assertion that human manipulation is neither a necessary nor sufficient condition of causality. However, pure isolation is impossibility. The principle of isolation assumes that χ1 and γ1 are free from all other influences or, as Malhotra (1996) puts it, that there is an absence of other causal factors. This implies that the two variables exist in a vacuum that excludes all other influences. In reality, phenomena that these variables represent are part of a complex of characteristics of individuals, groups, or other objects of study. The γ1 variable cannot happen in isolation since the units of analysis possess many characteristics besides χ1 on which they differ, and a number of these are expected to have some influence on γ1. Therefore, it cannot be stated with certainty that χ1 causes γ1.
The condition of association (concomitant variation) refers to the extent to which a cause, χ1, and an effect, γ1, vary together as predicted by the theory. In this respect, the common practice of assuming a bivariate relation between measured and latent variables is not fully justified. For example, consider the following item taken from the present instrument: “Apart from what we learned from the results, doing the study was educational”. Respondents are asked to indicate the level of their agreement with this statement. The resulting responses form the measured (observed) variable. However, a number of latent variables may underlie the level of agreement with this statement. One latent variable may be the level of respondents' own involvement with the research project. Those who had a direct involvement with the project, or those who had a contribution at the initiation of the project, are more likely to agree with this statement. The general attitude of respondents towards the usefulness of the particular type of project is also likely to influence their response to the statement. Utilisation as a latent variable may have a causal effect on the observed variable here, but it is likely that other latent variables also have an effect. A similar situation exists for many other indicators. In these situations, a bivariate association is neither necessary nor sufficient for a causal relationship between them.
Bollen (1989, p. 58) expresses this phenomenon in statistical terms:
Even under ideal conditions there are some complications to establishing the association. To illustrate, consider [the following]:
γ2 = β21 γ1 + γ21 χ1 + ζ2
The γ1 coefficient gives the association of χ1 and γ1. The γ11 is a
population parameter. We have ŷ11, a consistent estimator of γ11, as
the basis of making statements about γ11. In any given sample the
value of ŷ11 differs from γ11 because of sampling error. Usually we
can estimate the probability that γ11 takes particular values. But
these are probabilities not certainties, and mistakes in judging associations will occur. However, in practical terms we are usually willing to live with sampling error as long as we know its magnitude.
The third issue about causal relationships is the direction of causation. The likelihood of an association being causal is dependent on getting the direction of association right. “The time order of occurrence condition states that the causing event must occur either before or simultaneously with the effect; it cannot occur afterwards” (Malhotra 1996, p. 236). It is of course easier to claim a cause-effect relationship if there is a measurable time lag between the cause variable and the effect variable. To use an example from the present study if the decision-maker’s appreciation of the organisation’s past marketing research activity is positive then this would have a positive influence on the evaluation of the current marketing research project. This proposition makes intuitive sense and it is supported by substantive and theoretical work on organisational strategy in the past.
However, it is not known when the planning for the current marketing research project had started. If it is assumed that the decision-maker had some vested interest in the success of the current marketing research project, then her desire to have it approved might have influenced her part in the evaluation of past activity. In other words, it can be said (erroneously) that a future event partly caused a present one. In fact, the future is not the cause. It is the expectation of the approval of the later project that is the cause, not the later project itself, which had not happened yet. Nevertheless, the direction of the cause-effect relationship between the appreciation of the past activity and the evaluation of the most recent marketing research project is already in doubt. For this reason, although most of the causal relationships depicted in the model are not bi-directional, the cyclical nature of the marketing research activity and its consequences is continually asserted.